AI Rainfall Forecast for Harvest

I want to build a small AI-driven agent that concentrates purely on weather predictions, with rainfall and precipitation as its only inputs, so I can decide exactly when to harvest.

Here is what I need the system to do for me:
• Pull historical and real-time rainfall data from reliable public APIs (NOAA, Meteostat, or a source you recommend).
• Train a lightweight model in Python—TensorFlow, PyTorch, or an equally proven library—that forecasts field-level precipitation a few days ahead.
• Convert the raw forecast into a simple recommendation: safe-to-harvest, harvest-soon, or delay.
• Present that recommendation in a clear format. A small web dashboard is ideal, but a command-line script with a clean JSON output is acceptable if it speeds things up.
• Allow me to adjust key thresholds (e.g., maximum mm of expected rain) without touching code.

Acceptance criteria
• Forecasts must cover at least a 72-hour horizon and update automatically every six hours.
• The model should outperform a basic persistence baseline by a noticeable margin.
• Code, data-source references, and a brief README are delivered in a Git repository I can pull and run on Ubuntu.

If you can get a minimal version running quickly, we can iterate on additional features such as stock or waste tracking later, but for now my only target is precise rainfall-based harvest timing.